Question answering apparatus, question answering method, and program

Abstract

To provide a question answering device, a question answering method, and a program that enable a questioner to have an experience as if the questioner were actually asking a question to a specific person in mind.SOLUTION: The question answering device 10 includes an knowledge learning instructing unit 13, a manner learning instructing unit 14, and an answer instructing unit 16. The knowledge learning instruction unit 13 instructs the AI device 20 to learn knowledge included in the training data P1 created by the assumed responder D10. The style learning instruction unit 14 instructs the AI device 20 to learn, as a style, a component in at least one of the writing style and the speaking style of the expected responder D10 specified by the training data P1. The answer instruction unit 16 outputs the question from the questioner P2 to the AI device 20, thereby causing the AI device 20 to output an answer based on the knowledge and manner of the assumed answerer P1.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a question and answering device, a question and answering method, and a program. [Background technology]

[0002] Chatbots and the like are known as devices that automatically reply to input messages (see, for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7370115 specification Summary of the Invention [Problem to be solved by the invention]

[0004] For example, there are cases where a client who has contracted with a lawyer as a legal advisor wants to ask the lawyer a question. In such cases, the client's schedule may not match the lawyer's, or the client may hesitate to contact the lawyer because the question is minor. Even in such cases, if the client can have an experience that feels like asking the lawyer directly, the client's satisfaction with the lawyer can be increased.

[0005] A similar problem exists when a questioner wants to ask a specific person other than a lawyer a question, but finds it difficult to ask that person.

[0006] The present invention has been made in consideration of the above-mentioned problems, and aims to provide a question and answering device, a question and answering method, and a program that allow a questioner to have the experience of actually asking a question to a specific person in mind. [Means for solving the problem]

[0007] The present invention relates to the following question and answering device, question and answering method, and program.

[0008] (1) A knowledge learning instruction unit that instructs the AI ​​to learn knowledge contained in the training data created by the expected respondent; a manners learning instruction unit that instructs the AI ​​to learn, as manners, components of at least one of the writing style and the speaking style of the expected respondent identified in the training data; an answer instruction unit that outputs a question from a questioner to the AI, and causes the AI ​​to output an answer based on the knowledge of the expected answerer and the manners; A question and answering device comprising:

[0009] (2) The components include at least one of dialect and standard language, sentence structure, parts of speech, and honorific language; the sentence structure includes at least one of a subject, a predicate, a complement, a modifier, a conjunction, and an independent word; the parts of speech include at least one of a noun, a verb, an adjective, an adjectival verb, a particle, an auxiliary verb, an adverb, a conjunction, a conjunction, and an interjection; The honorific language includes at least one of honorific language, humble language, and polite language. The question and answering device according to claim 1.

[0010] (3) The question answering device according to (1) or (2), wherein, when the components include a dialect, the answer instruction unit instructs the AI ​​to adjust the degree of dialect in the answer based on the relationship between the region where the dialect is used and the region of the questioner.

[0011] (4) The question and answering device according to any one of (1) to (3), wherein the knowledge learning instruction unit instructs the AI ​​to prohibit learning of knowledge from data other than that permitted by the expected answerer.

[0012] (5) The question and answering device described in any one of (1) to (4), wherein the answer instruction unit instructs the AI ​​to output an answer including at least one of empathy for the questioner's feelings and encouragement when the AI ​​determines that it is better to include at least one of the empathy and encouragement.

[0013] (6) The question and answering device according to any one of (1) to (5), wherein the knowledge learning instruction unit is configured to output legal questions and answers to the AI ​​as the training data.

[0014] (7) The question answering device described in (6), wherein the answer instruction unit instructs the AI ​​to relatively reduce the ambiguity in the answer if the question is in a legal field in which the expected answerer is skilled, and instructs the AI ​​to relatively increase the ambiguity in the answer if the question is in a legal field other than the expected answerer's skilled.

[0015] (8) A knowledge learning instruction step instructing the AI ​​to learn knowledge contained in the training data created by the anticipated respondent; a manners learning instruction step of instructing the AI ​​to learn, as manners, components of at least one of the writing style and the speaking style of the expected respondent identified in the training data; an answer instruction step of outputting a question from a questioner to the AI, and causing the AI ​​to output an answer based on the knowledge of the expected answerer and the manners; A question-answering method comprising:

[0016] (9) A knowledge learning instruction step instructing the AI ​​to learn knowledge contained in the training data created by the expected respondent; a manners learning instruction step of instructing the AI ​​to learn, as manners, components of at least one of the writing style and the speaking style of the expected respondent identified in the training data; an answer instruction step of outputting a question from a questioner to the AI, and causing the AI ​​to output an answer based on the knowledge of the expected answerer and the manners; A program that causes a computer to execute the following. [Effects of the Invention]

[0017] According to the present invention, the questioner can have an experience that makes him / her feel as if he / she is actually asking a question to a specific person in mind. [Brief explanation of the drawings]

[0018] [Figure 1] FIG. 1 is a diagram illustrating an example of a question and answering system including a question and answering device. [Figure 2] FIG. 2 is a diagram showing an example of an expected respondent registration screen displayed on the screen of an expected respondent terminal by the setting unit of the question and answering device. [Figure 3] FIG. 3 is a diagram illustrating an example of expected respondent data. [Figure 4] FIG. 4 shows an example of a teacher data registration screen displayed on the screen of the expected answerer terminal by the setting unit of the question and answering device. [Figure 5] FIG. 5 is a diagram showing an example of training data for each expected respondent. [Figure 6] FIG. 6 is a diagram showing an example of a questioner registration screen displayed on the screen of the questioner terminal by the setting unit of the question and answering device. [Figure 7] FIG. 7 is a diagram illustrating an example of questioner data. [Figure 8] FIG. 8 is a diagram showing an example of a knowledge learning instruction sentence that the knowledge learning instruction unit gives to the AI ​​model. [Figure 9] FIG. 9 is a diagram showing an example of a manners learning instruction sentence that the manners learning instruction unit gives to the AI ​​model. [Figure 10] 10A and 10B are diagrams showing an example of a pre-learning screen displayed by the pre-learning instruction unit on the expected respondent terminal. [Figure 11] FIG. 11 is a diagram showing an example of a question and answer screen displayed on the questioner terminal. [Figure 12] FIG. 12 is a diagram showing an example of a question answering instruction sentence issued from the question answering device to the AI ​​model. [Figure 13] FIG. 13 is a diagram showing an example of the feedback screen. [Figure 14] FIG. 14 is a flowchart showing an example of the flow of the initial setting process. [Figure 15] FIG. 15 is a flowchart illustrating an example of the flow of the pre-learning process. [Figure 16] FIG. 16 is a flowchart illustrating an example of the flow of the pre-learning process. [Figure 17] FIG. 17 is a flowchart illustrating an example of the flow of processing by the question and answering device to answer a question from a questioner. [Figure 18] FIG. 18 is a flowchart illustrating an example of the flow of the feedback process. [Figure 19] FIG. 19 is a block diagram showing an example of a computer that realizes the question and answering device and the AI ​​device (AI model). DETAILED DESCRIPTION OF THE INVENTION

[0019] [Question Answering Device Overview] Fig. 1 is a diagram showing an example of a question and answering system 1 including a question and answering device 10. As shown in Fig. 1, question and answering device 10 includes a knowledge learning instruction unit 13 that uses the output of an expected answerer P1 as teacher data D10 and instructs an AI (AI device 20) to learn the knowledge included in the teacher data D10, and a manners learning instruction unit 14 that instructs AI device 20 to learn, as manners, components of at least one of the writing style and speaking style of expected answerer P1 identified in teacher data D10. Question and answering device 10 further includes an answer instruction unit 16 that outputs a question Q from a questioner P2 to AI device 20, thereby causing AI device 20 to output an answer A based on the knowledge and manners of expected answerer P1.

[0020] With the above configuration, the questioner P2 can obtain, through the question and answering device 10, an answer A based on the knowledge of the expected answerer P1, in the writing and speaking style of the expected answerer P1. This allows the questioner P2 to have the experience of actually asking the expected answerer P1 a question. In other words, the questioner P2 can obtain answer A in an atmosphere as if it were written or spoken by the expected answerer P1.

[0021] Next, an example of the configuration of the question and answering device 10 etc. will be described more specifically.

[0022] [Overall configuration of the question answering system] Question and answering system 1 is equipped with AI model 21 of AI device 20 that has learned from training data D10 prepared in advance by expected answerer P1. Question and answering system 1 answers question Q from questioner P2 by having AI model 21 generate answer A for question Q from questioner terminal 40 (questioner P2) and outputting this answer A to questioner terminal 40.

[0023] In this embodiment, an example will be described in which the expected respondent P1 is a lawyer and the questioner P2 is a client of the expected respondent P1. That is, this embodiment illustrates a configuration in which, even if the lawyer cannot directly communicate with the client, the client can ask the legal advisor a question as if they were conversing with the legal advisor.

[0024] In this embodiment, the question and answering device 10 is configured to perform the following processes.

[0025] (Overview of processing by the question answering device) The AI ​​device 20 is made to learn the training data D10 created by the expected respondent P1, thereby causing the AI ​​device 20 to generate an AI model 21 for each expected respondent P1. The AI ​​model 21 learns the knowledge contained in the training data D10 and also learns, as etiquette, components of at least one of the writing style and speaking style of the expected respondent P1 identified in the training data D10. The AI ​​model 21 answers the question Q received from the questioner P2, and outputs the answer A to the questioner terminal 40.

[0026] (More details about the processing by the question answering device) The AI ​​model 21 is made to perform pre-learning using the training data D10 in a manner involving the expected respondent P1. · The AI ​​model 21 provides answer A taking into consideration the distance between the area of ​​the questioner P2 and the area of ​​the expected respondent P1. · The AI ​​model 21 is prohibited from learning knowledge from data other than that permitted by the expected respondent P1. · Have AI model 21 provide answer A that is sensitive to the feelings of questioner P2. Instruct AI model 21 to give a clearer answer A for the legal field in which expected respondent P1 excels. The training data D10 is divided into training data common to all (multiple) questioners P2 and training data specific to a specific questioner P2, and the training data is trained by the AI ​​model 21. · The content of question Q and a record of answer A to question Q will be provided to expected respondent P1. The expected answerer P1 decides how to reflect the actual question Q from the questioner P2 in the AI ​​model 21.

[0027] (Question and Answer System) As shown in FIG. 1, the question and answering system 1 includes a question and answering device 10, an AI device 20, an assumed answerer terminal 30, and a questioner terminal 40.

[0028] The assumed answerer terminal 30 is operated by the assumed answerer P1. The assumed answerer terminal 30 executes a program provided by the question answering device 10, thereby becoming able to communicate with the question answering device 10. The assumed answerer terminal 30 accepts input from the assumed answerer P1 while the program is running. The assumed answerer terminal 30 accepts assumed answerer data D1 set by the assumed answerer P1. The assumed answerer terminal 30 outputs this assumed answerer data D1 as well as teacher data D10 input by the assumed answerer P1 to the question answering device 10. The assumed answerer terminal 30 also receives questioner data D2 and feedback information from the question answering device 10. The assumed answerer terminal 30 is connected to the question answering device 10 and the like via a network such as the Internet 5. Examples of the assumed answerer terminal 30 include a PC (personal computer), a smartphone, a tablet, a feature phone, smart glasses, and smart goggles.

[0029] The questioner terminal 40 is operated by the questioner P2. The questioner terminal 40 executes a program provided by the question and answering apparatus 10, thereby becoming capable of communicating with the question and answering apparatus 10. With the program executed, the questioner terminal 40 accepts input from the questioner P2. The questioner terminal 40 accepts questioner data D2 set by the questioner P2. The questioner terminal 40 outputs the question Q input by the questioner P2, in addition to the questioner data D2, to the question and answering apparatus 10. The questioner terminal 40 is connected to the question and answering apparatus 10 via a network such as the Internet 5. Examples of the questioner terminal 40 include a PC (personal computer), a smartphone, a tablet, a feature phone, smart glasses, and smart goggles.

[0030] AI device 20 is equipped with an AI model 21. Based on information (teacher data D10 and instructions (prompt)) provided by question answering device 10, AI model 21 generates an answer A to question Q provided by question answering device 10, and outputs this answer A to question answering device 10. Also, based on the information provided by question answering device 10, AI model 21 generates feedback information to be provided to expected answerer P1, and outputs this feedback information to question answering device 10. AI model 21 is a generative AI (Artificial Intelligence). AI model 21 is equipped with large language models (LLM).

[0031] For example, a large-scale language model performs tasks such as tokenizing an input sentence to break it down into its smallest units, understanding the context by calculating the relevance of each token in the prompt, extracting features (encoding), predicting the next token (decoding), and outputting the probability of the next token in the input sentence.

[0032] Specific examples of configurations equipped with large-scale language models include ChatGPT (registered trademark), PaLM, LaMDA, LLaMa, Claude (registered trademark), OpenCALM, Alpaca, etc. AI model 21 is only required to be able to output a request from question answering apparatus 10 to question answering apparatus 10 in the form requested by question answering apparatus 10, and the specific configuration is not limited.

[0033] (Anticipated respondent data) Fig. 2 is a diagram showing an example of an assumed answerer registration screen G1 displayed on the screen of the assumed answerer terminal 30 by the setting unit 12 of the question answering device 10. Fig. 3 is a diagram showing an example of assumed answerer data D1. As shown in Figs. 1 to 3, the assumed answerer registration screen G1 is a screen on which the question answering device 10 acquires the assumed answerer data D1. The question answering device 10 is configured to acquire one assumed answerer data D1 for each assumed answerer P1 (assumed answerer ID). The assumed answerer ID is assigned by the question answering device 10.

[0034] The expected respondent registration screen G1 displays, for example, an input field G11 for the expected respondent data D1. The expected respondent P1 can create the expected respondent data D1 by entering information in this input field G11. The expected respondent P1 may input each item in the expected respondent data D1 arbitrarily by operating a keyboard or the like, or may input each item by checking a preset item.

[0035] The expected respondent data D1 is data for identifying the type of person expected respondent P1 is. Examples of items that identify expected respondent P1 include expected respondent ID, name, qualifications, affiliation, base, and legal field of expertise. Qualifications indicate the qualifications held by expected respondent P1, and examples include national qualifications such as lawyer or tax accountant, public qualifications such as the Japan Chamber of Commerce and Industry Bookkeeping Examination, and other private qualifications. Affiliation can be the company or organization to which expected respondent P1 belongs. Examples of areas of expertise can be fields related to the qualifications, and in the case of a lawyer, the laws that the lawyer handles can be an example.

[0036] (Teacher data) In this embodiment, a trained model is used that is obtained by AI model 21 pre-training using a large-scale language model and teacher data D10 created by expected answerer P1. This AI model 21 receives question Q given to question answering device 10 from questioner terminal 40, generates answer A, and outputs this answer A to questioner terminal 40 via question answering device 10.

[0037] In this embodiment, the training data D10 includes a question Q and an answer A related to law, and this training data D10 is output to the AI ​​model 21.

[0038] FIG. 4 is a diagram showing an example of the teacher data registration screen G2 displayed on the screen of the expected respondent terminal 30 by the setting unit 12 of the question and answer device 10. FIG. 5 is a diagram showing an example of teacher data D10 for each expected respondent P1 (for each expected respondent ID). As shown in FIGS. 1, 4, and 5, teacher data D10 is created for each expected respondent P1 (for each expected respondent ID). One question and answer ID is assigned to each question and answer in the teacher data D10. The teacher data registration screen G2 displays a field G21 for specifying whether or not a questioner is limited to one question and answer, a question field G22, an answer field G23, and an input field G24 for the legal field primarily related to the question and answer. The expected respondent P1 can register questions and answers for each question and answer ID by entering information in these input fields G21 to G24. Each item in the teacher data D10 may be entered arbitrarily by the expected respondent P1 operating a keyboard or the like, or may be entered by loading a data file containing pre-recorded questions and answers into the expected respondent terminal 30.

[0039] The questioner limitation field G21 is a field for setting whether or not to answer regardless of who the questioner P2 is when the questioner P2 inputs the question Q into the questioner terminal 40 assuming the expected answerer P1. If, for example, "No limitation" is set in this field G21, the question and answering device 10 sets the content of the answer for the question and answer ID to be answered regardless of the questioner P2. On the other hand, if, for example, "Questioner X limitation" (X is any character or number) is set in this field G21, the question and answering device 10 sets the content of the answer for the question and answer ID to be answered only by the set specific questioner P2, and it is set that the answer A will not be used for answers A to questions Q from anyone other than the specific questioner P2.

[0040] The question field G22 is entered with a question Q created by the expected respondent P1. The answer field G23 is entered with an answer A created by the expected respondent P1. This answer A may be created by the expected respondent P1 operating a keyboard or the like, or by voice input using a voice input function provided in the expected respondent terminal 30 or the like. In the case of voice input, the voice input result is converted into text in the expected respondent terminal 30. The answer A by the expected respondent P1 may be created in written language or in colloquial language. The primarily related legal field field G24 is entered with laws that are primarily related to the content entered in the question field G22 and the answer field G23. The question Q and answer A output by the expected respondent P1 are included in the training data D10.

[0041] (Questioner data) Fig. 6 is a diagram showing an example of a questioner registration screen G3 displayed on the screen of the questioner terminal 40 by the setting unit 12 of the question and answering device 10. Fig. 7 is a diagram showing an example of questioner data D2. As shown in Figs. 1, 6 and 7, the questioner registration screen G3 is a screen on which the question and answering device 10 acquires the questioner data D2. The question and answering device 10 is configured to acquire the questioner data D2 for each questioner P2 (for each questioner ID). The question and answering device 10 assigns the questioner ID.

[0042] The questioner registration screen G3 displays an input field G31 for inputting questioner data D2. The questioner P2 can output the questioner data D2 to the question and answering device 10 by inputting information in this input field G31. Each item in the questioner data D2 may be input arbitrarily by the questioner P2 operating a keyboard or the like, or may be input by checking preset items. The questioner data D2 may be created by the expected answerer P1 instead of the questioner P2.

[0043] The questioner data D2 is data for identifying the type of person the questioner P2 is and for identifying which expected respondent P1 the questioner P2 desires. Examples of items for identifying the questioner P2 include name, affiliation, and base location. Examples of affiliation include the company or organization to which the questioner P2 belongs. Examples of items for identifying the expected respondent P1 desired by the questioner P2 include the expected respondent ID and the expected respondent name. The expected respondent ID and the expected respondent name are entered as the expected respondent ID and the expected respondent name in the expected respondent data D1. The dialect preference field is a field for selecting whether the questioner P2 desires an answer in the dialect used by the expected respondent P1 when asking question Q using the question and answering device 10 and obtaining answer A. The dialect preference field can be specified, for example, by operating a check box. Specifying a circle in the dialect preference field means that the questioner desires an answer in the dialect, while specifying an × means that the questioner does not desire an answer in the dialect.

[0044] (Specific Configuration of Question Answering Device) The question and answering device 10 is, for example, a server, and includes a calculation device such as a CPU (Central Processing Unit), a volatile memory such as a RAM (Random Access Memory), a non-volatile memory such as a ROM (Read Only Memory), and a storage device such as a hard disk or SSD (Solid State Drive). Data related to the question and answering device 10 may be stored in the question and answering device 10 or in an external storage device. The question and answering device 10 performs processing for each expected answerer P1.

[0045] The question and answering device 10 includes a receiving unit 11, a setting unit 12, a knowledge learning instruction unit 13, a manners learning instruction unit 14, a pre-learning instruction unit 15, an answer instruction unit 16, and a feedback unit 17.

[0046] The receiving unit 11 receives outputs from the AI ​​device 20, the assumed answerer terminal 30, and the questioner terminal 40.

[0047] The setting unit 12 performs settings in the question and answering device 10.

[0048] FIG. 8 is a diagram showing an example of a knowledge learning instruction statement R10 that the knowledge learning instruction unit 13 provides to the AI ​​model 21. While FIG. 8 and other figures illustrate an example of an explanation for expected respondent ID 1, similar processing is performed for other expected respondent IDs. As shown in FIGS. 1, 5, and 8, the knowledge learning instruction unit 13 uses the knowledge learning instruction statement R10 to instruct the AI ​​model 21 to learn knowledge contained in the training data D10. The knowledge in this case includes knowledge and insight, and in this embodiment, it particularly includes knowledge and insight related to law.

[0049] In this embodiment, the knowledge learning instruction sentence R10 instructs the AI ​​model 21 to prohibit learning of knowledge from data other than that permitted by the assumed respondent P1.

[0050] In this embodiment, the knowledge learning instruction sentence R10 limits the knowledge learning target at each of (i) the time of initial setup, (ii) the time of pre-learning, (iii) the time of question answering, and (iv) the time of feedback. (i) The time of initial setup refers to the time when the expected answerer P1, the questioner P2, and the teacher data D10 are registered in the question and answering device 10 and the AI ​​model 21. (ii) The time of pre-learning refers to the time when the AI ​​model 21 is caused to perform pre-learning using the teacher data D10, etc., prior to receiving a question Q from the questioner P2 so that the question and answering device 10 and the AI ​​model 21 can answer the question Q from the questioner P2. (iii) The time of question answering refers to the time when the question and answering device 10 and the AI ​​model 21 answer the question Q from the questioner P2 upon receiving the question Q from the questioner P2. (iv) The time of feedback refers to the time when the expected answerer P1 improves the AI ​​model 21 using the result of answering the question Q from the questioner P2.

[0051] The knowledge learning instruction sentence R10 includes instruction sentences R11 to R14 for (i) initial setup, (ii) pre-learning, (iii) answering questions, and (iv) feedback, respectively.

[0052] In addition, the knowledge learning instruction statement R10 may be a statement instructing the AI ​​model 21 to allow learning of knowledge other than those mentioned above, such as knowledge available on the Internet, that has been permitted by the expected respondent P1.

[0053] FIG. 9 is a diagram illustrating an example of an etiquette learning instruction sentence R20 provided to the AI ​​model 21 by the etiquette learning instruction unit 14. As illustrated in FIGS. 1, 5, and 9, the etiquette learning instruction unit 14 instructs the AI ​​model 21 to learn, as etiquette, components of at least one of the writing style and speaking style of the expected respondent P1 identified in the training data D10. The "writing style" refers to a written response by the expected respondent P1, and the "speaking style" refers to a colloquial response by the expected respondent P1. The components include at least one of dialect and standard language, sentence structure, parts of speech, and honorific language. Among these, the sentence structure includes at least one of a subject, a predicate, a complement, a modifier, a conjunction, and an independent word. The parts of speech include at least one of a noun, a verb, an adjective, an adjectival verb, a particle, an auxiliary verb, an adverb, a conjunction, a conjunction, and an interjection. The honorific language includes at least one of honorific language, humble language, and polite language.

[0054] The etiquette learning instruction R20 includes instructions R21 to R24 for (i) initial setup, (ii) pre-learning, (iii) question answering, and (iv) feedback. The etiquette learning instruction R20 also includes an instruction R25 that explains the components of etiquette. The etiquette learning instruction R20 may also include an instruction R26 that permits the use of information other than the training data D10, such as collective intelligence on the Internet. The use of such collective intelligence is performed so that the AI ​​model 21 can grasp the components (etiquette) contained in the training data D10, and is not intended to learn knowledge (such as legal knowledge or other substantive content of responses) from anyone other than the expected respondent P1.

[0055] The manners learning instruction unit 14 may instruct the AI ​​model 21 to prohibit learning of manners from anyone other than the expected answerer P1 (question answering system 1).

[0056] 10A and 10B are diagrams showing an example of a pre-learning screen G4 displayed by the pre-learning instructor 15 on the expected respondent terminal 30. As shown in FIGS. 1, 10A, and 10B, the pre-learning instructor 15 causes the AI ​​model 21 (large-scale language model) to perform pre-learning. While machine learning is shown as an example of the pre-learning method in this case, the specific learning method is not limited, and well-known learning methods such as deep learning, which is an example of machine learning, may also be used. For example, when the expected respondent P1 performs a predetermined operation on the expected respondent terminal 30 to output a pre-learning request to the question answering device 10, the pre-learning instructor 15 causes the expected respondent terminal 30 to display the pre-learning screen G4.

[0057] The pre-learning screen G4 is displayed so that the expected respondent P1 can perform pre-learning on the AI ​​model 21. Specifically, the pre-learning screen G4 is provided with an individual question input field G41 and a question batch conversion / input button G42. The individual question input field G41 is a field in which the expected respondent P1 can input characters arbitrarily, and is intended for the expected respondent P1 to input a hypothetical question Q that is an assumed question Q from the questioner P2.

[0058] Displayed to the side of the individual question input field G41 are a question as is button G43 and an AI-converted question button G44. The question as is button G43 is a button that issues a command to output the content entered in the individual question input field G41 as is from the expected answerer terminal 30 to the AI ​​model 21 via the question answering device 10. The AI-converted question button G44 is a button that issues an instruction to convert the content of the question Q entered in the individual question input field G41 by the AI ​​model 21 and answer it by the AI ​​model 21. When the AI-converted question button G44 is operated by the expected answerer P1, an instruction statement that instructs the content entered in the individual question input field G41 to be converted into a predetermined different expression is output from the question answering device 10 to the AI ​​model 21.

[0059] In this case, "changing to another expression" may mean (a) changing the content entered in the individual question input field G41 to another expression with the same meaning, or (b) changing the numerical value when the content entered in the individual question input field G41 includes a numerical value; the specific expression is not limited. In this embodiment, the pre-learning screen G4 may be displayed so that the above (a) and (b) can be individually specified.

[0060] When the question batch conversion / input button G42 is operated, an instruction statement instructing the AI ​​model 21 to change the question content to a predetermined different expression is output from the question answering device 10 to the AI ​​model 21 for each question and answer ID included in the training data D10. In this case, "changing to a different expression" may be the same as (a) or (b) above, and the specific expression is not limited. In this embodiment, the pre-learning screen G4 may be displayed so that the above (a) and (b) can be individually specified in relation to the question batch conversion / input button G42.

[0061] When any of the above buttons G42, G43, or G44 is operated on the pre-learning screen G4, the AI ​​model 21 performs processing in accordance with the provided instruction to generate an answer A. This answer A is provided to the expected answerer terminal 30 via the question answering device 10, and the pre-learning screen G5 is displayed on the expected answerer terminal 30, as shown in FIG. 10B. An answer field G51 is displayed on the pre-learning screen G5. When any of the buttons G43 or G44 related to the individual question input field G41 is operated, the answer A to the question Q is displayed in the answer field G51. On the other hand, when the question batch conversion / input button G42 is operated, the answer A to the question Q is displayed in the answer field G51 for each question and answer ID.

[0062] If the displayed answer A is suitable for presentation to the questioner P2, the expected answerer P1 operates the approval button G52 on the pre-learning screen G5, causing the AI ​​model 21 to learn that he or she has approved the answer A. On the other hand, if the displayed answer A is not suitable for presentation to the questioner P2 and the expected answerer P1 simply does not approve it, he or she operates the rejection button G53, causing the AI ​​model 21 to learn that the answer A is not suitable as an answer A to the questioner P2. Furthermore, if the expected answerer P1 feels that the displayed answer A is not suitable for presentation to the questioner P2 and wants to correct the answer A, he or she enters the correction content in the correction field G54 and operates the correction button G55. In this way, the question answering device 10 causes the AI ​​model 21 to learn that the answer before the correction was not suitable as an answer to the questioner P2, and causes the AI ​​model 21 to learn a preferable answer.

[0063] With the above configuration, an AI model 21 that has completed pre-learning is generated.

[0064] FIG. 11 is a diagram showing an example of the question and answer screen G6 displayed on the questioner terminal 40. FIG. 12 is a diagram showing an example of a question answering instruction sentence R30 issued from the question answering device 10 to the AI ​​model 21. As shown in FIGS. 1, 11, and 12, the question and answer screen G6 shows a question input field G61 and an answer display field G62. The questioner P2 inputs a question Q in the question input field G61. The input question Q is output to the question and answering device 10, and a question answering instruction sentence R30 including the input question Q is output from the question and answering device 10 to the AI ​​model 21. The question answering instruction sentence R30 is created by the answer instruction unit 16. The AI ​​model 21 then generates an answer A based on the question answering instruction sentence R30 and outputs this answer A to the question and answering device 10. The question and answering device 10 displays the answer A in the answer display field G62 of the question and answer screen G6. For question Q from questioner P2, the AI ​​model 21 for the expected answerer ID linked to the questioner P2 generates answer A. If the AI ​​model 21 cannot generate answer A because the knowledge accumulated in the AI ​​model 21 is insufficient for question Q, for example, it outputs answer A to the question answering device 10 indicating that the answer is unknown.

[0065] In this embodiment, the question Q and the answer A are also simply referred to as a Q&A. In the above Q&A, as described above, the answer instruction unit 16 outputs a question answer instruction sentence R30 including the question Q from the questioner P2 to the AI ​​model 21, thereby causing the AI ​​model 21 to output an answer A based on the knowledge and manners of the expected answerer P1.

[0066] The question and answer instruction sentence R30 includes instructions R31 to R39.

[0067] The instruction statement R31 includes the questioner ID of the questioner P2 who asked the question Q, the expected answerer ID linked to the questioner P2, and the question Q from the questioner P2.

[0068] The instruction sentence R32 includes a sentence that instructs the AI ​​model 21 to use knowledge in the AI ​​model 21 to respond by imitating the manner of the expected respondent P1 (at least one of writing style and speaking style).

[0069] Instruction statement R33 includes an instruction to respond in standard Japanese to questioner P2 (questioner ID) whose dialect preference is set to X, regardless of whether the learned data contains a dialect. Instruction statement R33 also includes an instruction to respond in dialect to questioner P2 (questioner ID) whose dialect preference is set to O, if the learned data contains a dialect.

[0070] In addition, instruction statement R34 includes a sentence that instructs questioner P2 (questioner ID) whose dialect preference is set to ○ to adjust the degree of dialect in answer A based on the relationship between the area where the dialect is used (the base location of expected answerer P1) and the area (base location) identified in questioner P2's regional information if the learned data contains a dialect.

[0071] Figure 12 shows an example of instruction statement R34, which instructs that if the distance between the base of expected respondent P1 and the base of questioner P2 is within a specified value (50 km), the dialects contained in answer A should be output without adjustment, and if the distance is more than 50 km, half of the dialects contained in answer A should be converted to standard Japanese (leaving at least one dialect).

[0072] Instruction R35 includes a sentence that instructs that the use of knowledge other than that contained in AI model 21 is prohibited when answering question Q.

[0073] The instruction statement R36 includes a sentence instructing that the Q&A limited to a specific questioner ID indicated in the teacher data D10 should only be used for the Q&A of that specific questioner ID, and a sentence instructing that the Q&A for the specific questioner ID should not be used for Q&As other than those of that specific questioner ID.

[0074] The instruction statement R37 includes an instruction sentence to determine the emotion of the questioner P2 from the question Q input by the questioner P2 at the questioner terminal 40. The AI ​​model 21 receives the instruction statement R37 and identifies the emotion of the questioner P2 from the question Q. A known configuration can be used as the configuration for identifying the emotion at this time. Since the process of identifying the emotion of the questioner P2 from the question Q is not a process of obtaining knowledge (legal knowledge) from a source other than the AI ​​model 21, data other than the learned data accumulated in the AI ​​model 21, such as the training data D10, may be used.

[0075] Instruction statement R38 is an instruction statement that, when the AI ​​model 21 determines that the questioner P2 is suffering from anxiety, worries, or the like, and when the AI ​​model 21 determines that the answer should include at least one of empathy and encouragement in response to the questioner P2's feelings, the AI ​​model 21 outputs answer A including at least one of empathy and encouragement. For example, when the question Q from the questioner P2 includes a phrase indicating confusion or anxiety, such as "I'm in trouble" or "I'm anxious," the AI ​​model 21 may determine that the questioner P2 is suffering from anxiety, worries, or the like. An example of an empathetic sentence in this case is a sentence such as "That's tough." Furthermore, an example of an encouraging sentence in this case is a sentence such as "I'm rooting for you."

[0076] Instruction sentence R39 includes a sentence instructing the AI ​​model 21 to relatively reduce the ambiguity in answer A if question Q is in a legal field in which expected respondent P1 is skilled. Instruction sentence R39 also includes a sentence instructing the AI ​​model 21 to relatively increase the ambiguity in answer A if question Q is in a legal field other than a legal field in which expected respondent P1 is skilled.

[0077] The instruction statement R39 includes an instruction to instruct the AI ​​model 21 to compare the "area of ​​expertise in law" described in the expected respondent data D1 with the law described in the "mainly related area of ​​law" column for answer A in the training data D10. This causes the AI ​​model 21 to determine whether the main area of ​​law for question Q is included in the area of ​​expertise in the expected respondent data D1. If the main area of ​​law for question Q is included in the area of ​​expertise in the expected respondent data D1, the AI ​​model 21 generates an ending for answer A, for example, "It is highly likely that...," thereby relatively reducing the ambiguity of answer A. On the other hand, if the main area of ​​law for question Q is not included in the area of ​​expertise in the expected respondent data D1, the AI ​​model 21 generates an ending for answer A, for example, "It is also possible that...," thereby relatively increasing the ambiguity of answer A.

[0078] 3 and 5, when the AI ​​model 21 receives a question Q related to question and answer ID 1 in the training data D10, which includes a legal field (civil law (property law)) in which the expected respondent P1 (e.g., expected respondent ID = 1) is skilled, the ambiguity of the answer A is made relatively small. On the other hand, when the AI ​​model 21 receives a question Q (a question about medical law) related to question and answer ID 3 in the training data D10, which does not include a legal field in which the expected respondent P1 (e.g., expected respondent ID = 1) is skilled, the ambiguity of the answer A is made relatively large.

[0079] 13 is a diagram showing an example of a feedback screen G7. As shown in FIGS. 1 and 13, the feedback unit 17 associates a question Q from the questioner P2 with an answer A from the AI ​​model 21 to the question Q and stores them. The feedback unit 17 instructs the AI ​​model 21 to self-learn by providing the AI ​​model 21 with an instruction statement to learn these Q&As and reflect the results in the AI ​​model 21. Furthermore, when the expected answerer P1 operates the expected answerer terminal 30 to accept a feedback display request, the feedback unit 17 displays the feedback screen G7 on the expected answerer terminal 30.

[0080] The feedback screen G7 displays, for example, a list of questions Q from each questioner P2 and answers A to the questions Q. When an answer A to one question Q from the questioner P2 is output from the AI ​​model 21 and then the questioner P2 asks a follow-up question in response to the answer A, the feedback unit 17 may display this series of Q&As as a single group. The follow-up question Q can be considered to be important to the questioner P2, or the answer A can be considered difficult to understand. Therefore, by highlighting these Q&As as a group, the efficiency of feedback by the expected answerer P1 can be improved.

[0081] The feedback unit 17 is also configured to be able to correct the answer A presented to the questioner P2. Specifically, on the feedback screen G7, for example, a correction button G72 is displayed next to the past Q&A column G71. When the expected answerer P1 operates the correction button G72, the expected answerer P1 can correct the content written in the past Q&A column G71 by operating the expected answerer terminal 30. The feedback unit 17 instructs the AI ​​model 21 to learn the corrected content.

[0082] Furthermore, the feedback unit 17 is configured to be able to delete the answer A presented to the questioner P2. Specifically, on the feedback screen G7, for example, a delete button G73 is displayed next to the past Q&A column G71. When the expected answerer P1 operates the delete button G73, the feedback unit 17 instructs the AI ​​model 21 to delete the Q&A from the AI ​​model 21. For example, if the answer A to the question Q of the questioner P2 is inconsistent, it is possible to prevent the answer A from having an unnecessary influence on the answers A to future questions Q.

[0083] The above is a schematic configuration of the question and answering system 1 including the question and answering device 10. Next, an example of the flow of operations in the question and answering system 1 (question and answering device 10) will be described.

[0084] [An example of the operation flow of a question-answering system] 14 to 18 are flowcharts showing an example of processing in the question answering device 10. In the following description, FIGS. 1 to 18 will be referred to as appropriate. In addition, in this embodiment, a question answering method is implemented by operating the question answering device 10. Therefore, the description of the question answering method in this embodiment will be replaced by the following description of the operations.

[0085] (An example of the initial setting process flow) 14 is a flowchart showing an example of the flow of the initial setting process. In the initial setting, the expected respondent data D1 created by the expected respondent P1 operating the expected respondent terminal 30 is received by the receiving unit 11 (step S11). Furthermore, the teacher data D10 created by the expected respondent P1 is received by the receiving unit 11 through the expected respondent terminal 30 (step S12). Furthermore, the questioner data D2 created by the questioner P2 operating the questioner terminal 40 or the like is received by the receiving unit 11 (step S13). Note that the order in which steps S11 to S13 are performed does not have to be as described above, and steps S11 to S13 may be performed in an order different from the above.

[0086] The question answering device 10 outputs the received expected respondent data D1, teacher data D10, and questioner data D2 to the AI ​​model 21 (step S14). Next, the knowledge learning instruction unit 13 outputs a knowledge learning instruction statement R10 to the AI ​​model 21, instructing the AI ​​model 21 to learn the knowledge, etc. included in the teacher data D10 (step S15). Furthermore, the manners learning instruction unit 14 outputs a manners learning instruction statement R20 to the AI ​​model 21, instructing the AI ​​model 21 to learn, as manners, components of at least one of the writing style and speaking style of the expected respondent P1 identified in the teacher data D10 (step S16). The order of these steps S15 and S16 may be reversed. The AI ​​model 21 learns this knowledge and manners and becomes an AI model 21 for each expected respondent P1 (each expected respondent ID).

[0087] (An example of the pre-learning process flow) 15 and 16 are flowcharts for explaining an example of the flow of the pre-learning process. The pre-learning is started, for example, by the expected respondent P1 operating a pre-learning button (not shown) displayed on the expected respondent terminal 30. The pre-learning is performed for each expected respondent ID.

[0088] In the pre-learning, the pre-learning instruction unit 15 determines whether a question batch conversion request has been issued by operating the question batch conversion / input button G42 on the pre-learning screen G4 (step S21). If the question batch conversion / input button G42 has been operated (YES in step S21), an instruction statement instructing the AI ​​model 21 to convert the question content into a predetermined different expression is output from the question answering device 10 to the AI ​​model 21 for each question and answer ID included in the teacher data D10 (step S22), and the question content is converted into the predetermined different expression in the AI ​​model 21.

[0089] On the other hand, if the question batch conversion / input button G42 is not operated (NO in step S21) and the question Q is not input from the expected answerer P1 (expected answerer terminal 30) in the individual question input field G41 (NO in step S23), the question answering device 10 enters a standby state. On the other hand, if the expected answerer P1 inputs the question Q in the individual question input field G41 (YES in step S23) and then the post-AI converted question button G44 is operated (AI conversion in step S24), an instruction statement instructing the AI ​​model 21 to convert the content input in the individual question input field G41 into a predetermined different expression is output from the question answering device 10 to the AI ​​model 21 (step S25). As a result, the content input in the individual question input field G41 is converted into the predetermined different expression in the AI ​​model 21. On the other hand, when the expected answerer P1 inputs question Q into the individual question input field G41 (YES in step S23) and then operates the question button G43 (input as is in step S24), the content written in the individual question input field G41 is output as is from the question answering device 10 to the AI ​​model 21.

[0090] When AI model 21 receives question Q whose text has been converted in accordance with the instructions in step S22 or step S25, or question Q whose text was input in individual question input field G41 in step S24 without being converted, it generates answer A to question Q (step S26). Answer A is output from AI model 21 to expected answerer terminal 30 via question answering device 10, and is displayed in answer field G51 on pre-learning screen G5 on expected answerer terminal 30 (step S27).

[0091] When the expected respondent P1 operates the approve button G52 because the displayed answer A is suitable for presentation to the questioner P2 (approve in step S28), the pre-learning instruction unit 15 notifies the AI ​​model 21 that answer A is suitable (step S29) and causes the AI ​​model 21 to learn. On the other hand, when the expected respondent P1 operates the reject button G53 because the displayed answer A is not suitable for presentation to the questioner P2 (reject in step S28), the pre-learning instruction unit 15 notifies the AI ​​model 21 that answer A is rejected (step S30) and causes the AI ​​model 21 to learn. There may also be cases where the expected respondent P1 finds that the displayed answer A is not suitable for presentation to the questioner P2 and corrects answer A (correct in step S28). In this case, when the expected respondent P1 enters the correction content in the correction field G54 and operates the correction button G55, the correction instruction is accepted (step S31), and the pre-learning instruction unit 15 sends an instruction to the AI ​​model 21 to correct the answer A to the content of the correction field G54 (step S32), and causes the AI ​​model 21 to learn the correction content.

[0092] The pre-learning instructing unit 15 repeats the processes of steps S21 to S32 until an end instruction is received from the expected respondent terminal 30 by a predetermined operation by the expected respondent P1 (NO in step S33). On the other hand, when the pre-learning instructing unit 15 receives an end instruction from the expected respondent terminal 30 by a predetermined operation by the expected respondent P1 (YES in step S33), the pre-learning by the question answering device 10 and the AI ​​model 21 is terminated.

[0093] (An example of the question and answer process) 17 is a flowchart illustrating an example of the flow of processing by the question answering device 10 to answer a question Q from a questioner P2. This processing is executed when the questioner P2 starts up the question answering device 10. In this processing, when the questioner P2 operates the questioner terminal 40 to input question Q into the question input field G61 of the question and answer screen G6 and operates the question send button G63 displayed on the questioner terminal 40 (YES in step S41), the answer instruction unit 16 of the question answering device 10 creates a question answering instruction statement R30 including the question Q from the questioner P2, and outputs this instruction statement R30 to the AI ​​model 21 (step S42).

[0094] Upon receiving question answering instruction R30, AI model 21 generates answer A to question Q in accordance with question answering instruction R30 (step S43). Upon receiving answer A, question answering device 10 outputs this answer to questioner terminal 40 (step S44), and answer A is displayed on question and answer screen G6 on questioner terminal 40. If an additional question Q is input on questioner terminal 40 and question send button G63 is operated (YES in step S45), answer instruction unit 16 repeats the processing from step S42 onwards. On the other hand, if answer instruction unit 16 does not receive an additional question Q from questioner terminal 40 (NO in step S45), the processing for answering question Q ends.

[0095] (An example of the feedback process flow) 18 is a flowchart illustrating an example of the flow of the feedback process. The feedback process is started, for example, by expected respondent P1 operating a feedback button (not shown) displayed on expected respondent terminal 30 (YES in step S51). When the feedback button is operated, feedback unit 17 displays question Q from questioner P2 and answer A to question Q in past Q&A column G71, as shown on feedback screen G7 in FIG. 13 (step S52). These Q&As may be stored in question and answering device 10, or Q&As stored in AI model 21 may be displayed on questioner terminal 40.

[0096] When the feedback unit 17 receives a request for correction of the specified Q&A from the questioner terminal 40 (YES in step S53), and receives the correction content entered on the questioner terminal 40 and then receives the operation of the correction button G72 (step S54), it instructs the AI ​​model 21 to learn the corrected content, and causes the AI ​​model 21 to learn the corrected content (step S55).

[0097] On the other hand, when the feedback unit 17 receives that the delete button G73 has been operated for the specified Q&A on the questioner terminal 40 (YES in step S56) even though there is no request for correction (NO in step S53), it instructs the AI ​​model 21 to delete the Q&A and causes it to learn (step S57).

[0098] The feedback unit 17 repeats the processes of steps S52 to S58 until the end button G74 is operated (YES in step S58) on the expected respondent terminal 30. On the other hand, when the end button G74 on the feedback screen G7 is operated (YES in step S58), the feedback unit 17 ends the feedback process.

[0099] [Effects of the embodiment] As described above, according to this embodiment, the questioner P2 can obtain, through the question and answering device 10, an answer A based on the knowledge of the expected answerer P1, in the writing and speaking style of the expected answerer P1. This allows the questioner P2 to have the experience of actually asking the expected answerer P1 a question. In other words, the questioner P2 can obtain answer A in a way that sounds as if it were written or spoken by the expected answerer P1.

[0100] Furthermore, according to this embodiment, the way of answering question Q (answer format, style) can be adapted to the answering style of expected answerer P1, taking into consideration dialect, sentence structure, parts of speech, honorific language, etc. As a result, when questioner P2 asks a question, answer A can be output from question answering device 10 to questioner terminal 40 in a way of answering that is closer to that of expected answerer P1.

[0101] Furthermore, according to this embodiment, when answer A in teacher data D10 created by expected answerer P1 contains a dialect, answer instructing unit 16 instructs AI model 21 to adjust the degree of dialect in answer A based on the relationship between the region where the dialect is used and the region of questioner P2. With this configuration, the closer the base location of questioner P2 is to the base location of the expected answerer, the stronger the degree of dialect in answer A from question answering device 10 can be, and the questioner P2 can be given an interval during which he or she communicates more naturally with expected answerer P1.

[0102] Furthermore, according to this embodiment, the knowledge learning instruction unit 13 instructs the AI ​​model 21 to prohibit learning of knowledge from data other than that permitted by the expected respondent P1. This configuration can prevent the AI ​​model 21 from searching the Internet or the like on its own and creating an answer A that is not intended by the expected respondent P1.

[0103] Furthermore, according to this embodiment, when the AI ​​model 21 determines that it is better to include at least one of empathy and encouragement for the questioner P2's feelings, the answer instructing unit 16 instructs the AI ​​model 21 to output an answer A that includes at least one of empathy and encouragement. This configuration allows the questioner P2 to have the experience of the expected answerer P1 providing an answer to the questioner P2 and cheering him on. This further increases the expected answerer P1's trust in the questioner P2.

[0104] Furthermore, according to this embodiment, an environment can be realized in which legal professionals such as lawyers can provide legal consultation services to specific customers.

[0105] Furthermore, according to this embodiment, if the question Q is in a legal field in which the expected respondent P1 is skilled, the answer instruction unit 16 instructs the AI ​​model 21 to relatively reduce the ambiguity in the answer A, and if the question Q is in a legal field other than the expected respondent P1's skilled, the answer instruction unit 16 instructs the AI ​​model 21 to relatively increase the ambiguity in the answer A. With this configuration, a legal consultation service that takes into account the actual strengths of the expected respondent P1 can be provided to the asker P2.

[0106] [program] An example of the program in this embodiment is a program that causes a computer to execute steps S11 to S16, S21 to S25, S27 to S33, S41 to S42, and S44 to S45 shown in Figures 14 to 18. By installing and executing this program in a computer, question and answering device 10 and question and answering method can be realized. In this case, the processor of the computer functions as receiving unit 11, setting unit 12, knowledge learning instruction unit 13, manners learning instruction unit 14, advance learning instruction unit 15, answer instruction unit 16, and feedback unit 17 and performs processing.

[0107] The program in this embodiment may be executed by a computer system constructed by a plurality of computers, in which case, for example, each computer may function as one of the reception unit 11, setting unit 12, knowledge learning instruction unit 13, manners learning instruction unit 14, advance learning instruction unit 15, answer instruction unit 16, and feedback unit 17.

[0108] An example of the program in this embodiment is a program that causes a computer to execute steps S26 and S43 shown in Figures 15 and 17. By installing and executing this program in a computer, it is possible to realize AI device 20. In this case, the processor of the computer functions as AI model 21 and performs processing.

[0109] An example of a computer that executes the program of this embodiment to realize the question answering device 10 (question answering method) and the AI ​​device 20 will now be described with reference to Fig. 19. Fig. 19 is a block diagram showing an example of a computer that realizes the question answering device 10 and the AI ​​device 20 (AI model 21).

[0110] 19, a computer 110 includes a CPU (Central Processing Unit) 111, a main memory 112, a storage device 113, an input interface 114, a display controller 115, a data reader / writer 116, and a communication interface 117. These components are connected to each other via a bus 121 so as to be able to communicate data with each other.

[0111] Furthermore, the computer 110 may include a GPU (Graphics Processing Unit) or an FPGA (Field-Programmable Gate Array) in addition to or instead of the CPU 111. In this aspect, the GPU or FPGA can execute the programs in the embodiments.

[0112] The CPU 111 loads a program in the embodiment, which is composed of a group of codes and stored in the storage device 113, into the main memory 112 and executes each code in a predetermined order to perform various calculations. The main memory 112 is typically a volatile storage device such as a DRAM (Dynamic Random Access Memory).

[0113] The program in the embodiment may be provided in a state stored in computer-readable recording medium 120. The program in the embodiment may be distributed over the Internet connected via communication interface 117.

[0114] Specific examples of the storage device 113 include a hard disk drive and a semiconductor storage device such as a flash memory. The input interface 114 mediates data transmission between the CPU 111 and input devices 118 such as a keyboard and a mouse. The display controller 115 is connected to a display device 119 and controls the display on the display device 119.

[0115] Data reader / writer 116 mediates data transmission between CPU 111 and recording medium 120, reads programs from recording medium 120, and writes processing results from computer 110 to recording medium 120. Communication interface 117 mediates data transmission between CPU 111 and other computers.

[0116] Specific examples of the recording medium 120 include general-purpose semiconductor storage devices such as CF (Compact Flash (registered trademark)) and SD (Secure Digital), magnetic recording media such as flexible disks, or optical recording media such as CD-ROMs (Compact Disk Read Only Memory).

[0117] Note that the question and answering apparatus 10 in this embodiment can be realized not by a computer on which a program is installed, but by hardware corresponding to each unit, such as an electronic circuit. Furthermore, the question and answering apparatus 10 may be partially realized by a program and the remaining unit by hardware. In the embodiment, the computer is not limited to the computer shown in FIG. 19.

[0118] [Variations] The above describes an embodiment of the present invention. However, the present invention is not limited to the above embodiment. Various modifications of the present invention are possible within the scope of the claims. Below, configurations that differ from the above embodiment will be mainly described, and similar configurations will be denoted by similar reference numerals in the drawings and detailed description will be omitted.

[0119] In the above embodiment, various commands are given to the computer by the expected answerer P1 or the questioner P2 operating buttons, but this is not necessarily the case. For example, commands similar to those given by operating a button may be given to the computer by operating another screen display or by voice input, instead of button operation. Voice input may be used instead of character input, and voice output may be used instead of a screen display.

[0120] Furthermore, in the above embodiment, an example has been described in which the question answering device 10 communicates with the AI ​​device 20 via the Internet 5. However, this does not have to be the case. For example, the question answering device 10 and the AI ​​device 20 may be connected via a local area network (LAN), or an AI model having the same functions as the AI ​​model 21 of the AI ​​device 20 may be constructed on the same computer as the question answering device 10, and this AI model may be used as the AI ​​of the present invention.

[0121] Furthermore, in the above embodiment, an example has been described in which the instruction sentences from the question and answering apparatus 10 to the AI ​​model 21 are written in Japanese, but this does not have to be the case. The instruction sentences from the question and answering apparatus 10 to the AI ​​model 21 may be written in a language other than Japanese, such as English or a programming language.

[0122] Furthermore, in the above-described embodiment, the AI ​​model 21 does not need to be trained in advance.

[0123] In the above embodiment, legal consultation by a lawyer has been described as an example, but this is not the only option. For example, the present invention may be applied to legal consultation by a legal professional other than a lawyer. Furthermore, the present invention may be applied to cases where consultation services are provided to specific clients, not just lawyers. [Industrial Applicability]

[0124] The present invention can be applied as a question and answering device, a question and answering method, and a program. [Explanation of symbols]

[0125] 10 Question answering device 13 Knowledge Learning Instructions 14 Manners learning instruction section 16 Answer instructions section 21 AI Models (AI) D10 Training data P1 Expected respondents

Claims

1. A knowledge learning instruction unit that instructs the AI ​​to learn knowledge contained in the teacher data created by the expected respondent; a manners learning instruction unit that instructs the AI ​​to learn, as manners, components of at least one of the writing style and the speaking style of the expected respondent identified in the training data; An answer instruction unit that outputs a question from a questioner to the AI, and causes the AI ​​to output an answer based on the knowledge and manners of the expected answerer; A question and answering device comprising:

2. The components include at least one of dialect and standard language, sentence structure, parts of speech, and honorific language; the sentence structure includes at least one of a subject, a predicate, a complement, a modifier, a conjunction, and an independent word; the parts of speech include at least one of a noun, a verb, an adjective, an adjectival verb, a particle, an auxiliary verb, an adverb, a conjunction, a conjunction, and an interjection; The honorific language includes at least one of honorific language, humble language, and polite language. The question and answering device according to claim 1 .

3. 3. The question and answering device according to claim 2, wherein, when the components include a dialect, the answer instruction unit instructs the AI ​​to adjust the degree of the dialect in the answer based on a relationship between a region in which the dialect is used and a region of the questioner.

4. 2. The question and answering device according to claim 1, wherein the knowledge learning instruction unit instructs the AI ​​to prohibit learning of knowledge from data other than that permitted by the expected answerer.

5. 2. The question and answering device according to claim 1, wherein, when the AI ​​determines that it is better to include at least one of empathy for the questioner's feelings and encouragement, the answer instruction unit instructs the AI ​​to output an answer that includes at least one of the empathy and the encouragement.

6. 6. The question and answering device according to claim 1, wherein the knowledge learning instruction unit is configured to output legal questions and answers to the AI ​​as the training data.

7. 7. The question and answering device according to claim 6, wherein the answer instruction unit instructs the AI ​​to make the answer relatively less vague if the question is in a legal field in which the expected answerer is knowledgeable, and instructs the AI ​​to make the answer relatively more vague if the question is in a legal field other than the expected answerer's knowledge.

8. A knowledge learning instruction step of instructing the AI ​​to learn knowledge contained in the teacher data created by the expected respondent; an etiquette learning instruction step of instructing the AI ​​to learn, as etiquette, components of at least one of the writing style and the speaking style of the expected respondent identified in the training data; an answer instruction step of outputting a question from a questioner to the AI, and causing the AI ​​to output an answer based on the knowledge and manners of the expected answerer; A question-answering method comprising:

9. A knowledge learning instruction step of instructing the AI ​​to learn knowledge contained in the teacher data created by the expected respondent; an etiquette learning instruction step of instructing the AI ​​to learn, as etiquette, components of at least one of the writing style and the speaking style of the expected respondent identified in the training data; an answer instruction step of outputting a question from a questioner to the AI, and causing the AI ​​to output an answer based on the knowledge and manners of the expected answerer; A program that causes a computer to execute the following.

Images